A humanoid robot can share a floor with people only if its failure behaviour is engineered, verified and documented. Ars Technica reports that Agility Robotics’ new machine stops and squats; here is how to judge claims like that.
Key takeaways
- Ars Technica reports that Agility Robotics’ new humanoid robot is designed to stop and squat to avoid harming human coworkers, and that this opens the way to robots working outside physical cages and without safety barriers.
- A bipedal robot introduces a hazard that bolted-down industrial arms do not have, because a machine that can walk can also fall on somebody.
- Removing a safety fence is a risk-assessment decision about a whole installation, not a feature that a single robot can deliver on its own.
- Established machinery safety practice already describes fenceless operation through concepts such as monitored stops, speed and separation monitoring, and limits on contact force.
- The specifications, safety ratings and independent assessments of the robot described in the report are not detailed in the material available, and should be requested from the manufacturer rather than assumed.
What has actually been announced
Ars Technica reports that Agility Robotics’ new humanoid robot is built to stop and squat when continuing to move could harm a human working nearby, and that the consequence is that robots of this kind can begin operating outside physical cages and without safety barriers.
Two claims are worth separating. The first is the behaviour itself: a legged machine that lowers its centre of mass instead of simply cutting power and toppling. The second is the conclusion attached to it, that barrier-free operation becomes possible. The first is an engineering decision made by the manufacturer. The second is an outcome of risk assessment and regulation at each individual site, and it depends on far more than one movement.
Details of the machine — its mass, its sensing hardware, the integrity rating of its safety control system, and which independent bodies, if any, have assessed it — are not covered by the material available here. Those points should be treated as unknown rather than filled in by inference. Anyone weighing up a deployment will need that documentation from the supplier, and the rest of this guide sets out what to ask for and why each item matters.
Why cage-free operation is being discussed now
Humanoid and legged robots have moved from research demonstrations towards pilot work in logistics and manufacturing over the past few years. The commercial argument is straightforward: a machine shaped roughly like a person can, in principle, use stairs, aisles, shelves and trolleys that were designed for people, without rebuilding the building around it.
That argument collapses if the robot has to be fenced. A caged humanoid loses the flexibility that justified its human-like form in the first place, and occupies floor space that a conventional, cheaper, faster robot arm would use better. So the safety question is not a detail attached to the product — it is the product. Announcements about stopping behaviour, fall behaviour and contact forces are therefore the announcements that actually determine whether these machines have a business case.
This also explains why a squat is being presented as significant. It is a statement about what the robot does when something goes wrong, rather than about what it can achieve when everything goes right.
The background a newcomer needs
Traditional industrial robots are fast, strong and blind to their surroundings. The historic answer was separation: put the robot inside a fence, interlock the gate so opening it cuts power, and keep people out while it runs.
Collaborative operation replaced that blunt rule with a set of engineered alternatives. Machinery safety standards published by the International Organization for Standardization, together with related functional-safety standards, describe approaches including a safety-rated monitored stop, where the robot holds position while a person is present; speed and separation monitoring, where the robot slows or stops as the gap narrows; and power and force limiting, where the machine is designed so that any contact stays below a level judged capable of causing injury.
None of these is a property of the robot alone. Each depends on sensors, on the reliability of the control system that acts on them, on the task, on the payload, and on the layout of the cell. A robot can be certified as capable of a mode without the installation being safe to run in it.
Legged robots add a hazard that fixed arms do not have. A bolted-down arm that loses power stops where it is. A bipedal machine that loses balance becomes a falling mass. Squatting addresses that directly: lowering the body reduces the height a robot can fall from and the energy involved.
Who is affected, and how
Warehouse and production workers are affected most immediately, because they are the people who would share an aisle with such a machine. For them, the practical questions are about predictability: does the robot behave the same way every time, is its intention visible, and does it yield to a person rather than expecting the person to yield to it.
Health and safety managers and machinery integrators carry the legal weight. In most jurisdictions the employer or the integrator, not the robot manufacturer, is responsible for the safety of the installed system. A manufacturer’s claim is an input to that assessment, not a substitute for it.
Insurers, works councils and trade unions are affected in a slower way. Fencing is easy to inspect; a software-defined safety behaviour is not. Verifying that a robot still stops correctly after a firmware update is a different kind of task from checking that a gate interlock still works.
Facility designers are affected too. If barrier-free operation genuinely works, floor plans, fire routes and pedestrian walkways can be drawn differently. If it does not, the retrofit cost lands on them.
Where informed people disagree
There is broad agreement that fenceless robot operation is achievable, because it is already routine for some arms and for automated mobile robots in warehouses. The disagreement concerns legged humanoids specifically.
One camp argues that balance is a solved control problem in constrained indoor environments, that redundant sensing can detect people reliably, and that a falling robot is no more dangerous than the loaded trolleys and forklifts already moving through the same spaces.
The other camp argues that a humanoid has a higher centre of mass, more ways to fail, and a much larger behavioural state space than an arm following a fixed path — which makes exhaustive validation harder. Their concern is not whether the robot works, but whether anyone can demonstrate how often it will not.
A third line of disagreement is about standards themselves. Existing machinery standards were written with fixed arms and wheeled vehicles in mind, and there is ongoing debate about whether legged machines need their own requirements or fit within the current framework.
Six things to check before a barrier comes down
- Ask for the safety function list. Which specific functions are safety-rated, and to what performance level or safety integrity level? A general claim of safe behaviour is not an answer.
- Ask who verified it. Manufacturer self-declaration and third-party assessment are different things. Request the certificate or the test report, and note which one you are given.
- Separate the robot from the installation. Your own risk assessment covers the task, the payload, the floor surface, lighting, aisle width and the other traffic in the space.
- Test the failure cases, not the demo. What happens on power loss, sensor occlusion, network dropout, a slippery patch of floor, or a person appearing from behind a rack?
- Plan for change control. Decide in advance who re-validates safety behaviour after a software update, and what evidence is retained.
- Train the people who share the space. Fenceless operation transfers part of the safety burden to human judgement, and that only works if workers know what the robot will do.
What to watch next
Three things will show whether cage-free humanoid work is real or provisional. The first is published, independently verified safety documentation rather than demonstration video — specifically, evidence of which safety functions are rated and how. The second is the standards process: whether bodies working on machinery and robot safety issue requirements aimed at legged machines, and what they say about falls.
The third is deployment shape. If these robots appear only in areas that are still segregated by time, by shift or by soft boundaries, then the barrier has been moved rather than removed. Genuinely mixed operation, documented and audited, is the test that matters.
Frequently asked questions
Why would a robot squat instead of just switching off?
Cutting power to a bipedal machine does not make it safe, because a robot standing on two legs needs active control to stay upright. Ars Technica reports that Agility Robotics’ new humanoid is designed to stop and squat to avoid harming human coworkers. Lowering the body reduces the height from which the machine could fall and the energy of any impact, so it fails in a more controlled way.
Does this mean safety fences are no longer needed?
No. Fenceless operation is a conclusion reached through risk assessment of a complete installation, covering the task, the environment and the people involved, not a feature a robot can supply by itself. Machinery safety standards already describe how to operate without barriers, but they require engineered and verified safety functions. A manufacturer’s design choice is evidence for that assessment, not a replacement for it.
Are humanoid robots more dangerous than industrial robot arms?
They present different hazards. A bolted-down arm is typically faster and stronger within its reach but stops where it is when power is lost. A legged robot is generally slower, yet it can travel, it has a higher centre of mass, and it can fall. Comparative injury data for humanoid machines in real workplaces is not available in the material covered here.
Who is legally responsible if a fenceless robot injures someone?
In most jurisdictions, responsibility for the safety of an installed machine system rests with the employer and the integrator who put it into service, alongside the manufacturer’s duties for the product itself. That is why site-specific risk assessment, documented safety functions and worker training matter. Requirements vary by country, so the relevant national occupational safety regulator should be consulted directly.
What standards apply to robots working alongside people?
The core framework comes from International Organization for Standardization documents on industrial robot safety and collaborative operation, supported by functional-safety standards that define how reliable a safety function must be. These describe approaches such as safety-rated monitored stops, speed and separation monitoring, and limits on contact force. Whether legged humanoids need additional, purpose-written requirements is currently an open question in the standards community.
What should a worker ask before sharing a floor with one?
Ask what the robot does when it detects you, how you can tell it has seen you, where the emergency stops are and how to reach them, what the machine does on power or network loss, and which areas remain off limits. Ask who to report unexpected behaviour to. Predictability and clear reporting routes matter more than the robot’s peak performance figures.
Sources and further reading
- Ars Technica — technology news report describing the robot’s stop-and-squat behaviour and the prospect of barrier-free operation.
- International Organization for Standardization — published standards and technical specifications on industrial robot safety, collaborative operation and machinery risk assessment.
- National occupational safety and health regulators — official guidance on employer duties for machinery installations and workplace risk assessment.
- Robotics industry trade associations — explanatory material on collaborative robot modes, safety functions and integration practice.
Surfaced from the rss:arstechnica signal “humanoid robot safety design”. AI-assisted draft, editorially reviewed.
